Mark E. Olszewski

Automatic Model-Based Segmentation of the Heart in CT Images

Automatic image processing methods are a pre-requisite to efficiently analyze the large amount of image data produced by computed tomography (CT) scanners during cardiac exams. This paper introduces a model-based approach for the fully automatic segmentation of the whole heart (four chambers, myocardium, and great vessels) from 3-D CT images. Model adaptation is done by progressively increasing the degrees-of-freedom of the allowed deformations. This improves convergence as well as segmentation accuracy. The heart is first localized in the image using a 3-D implementation of the generalized Hough transform. Pose misalignment is corrected by matching the model to the image making use of a global similarity transformation. The complex initialization of the multicompartment mesh is then addressed by assigning an affine transformation to each anatomical region of the model. Finally, a deformable adaptation is performed to accurately match the boundaries of the patients anatomy. A mean surface-to-surface error of 0.82 mm was measured in a leave-one-out quantitative validation carried out on 28 images. Moreover, the piecewise affine transformation introduced for mesh initialization and adaptation shows better interphase and interpatient shape variability characterization than commonly used principal component analysis.

Coronary plaque imaging with 256-slice multidetector computed tomography: interobserver variability of volumetric lesion parameters with semiautomatic plaque analysis software

The purpose of this study was to evaluate the potential clinical value of coronary plaque imaging with a new generation CT scanner and the interobserver variability of coronary plaque assessment with a new semiautomatic plaque analysis application. Thirty-five isolated plaques of the left anterior descending coronary artery from 35 patients were evaluated with a new semiautomatic plaque analysis application. All patients were scanned with a 256-slice MDCT scanner (Brilliance iCT, Philips Healthcare, Cleveland OH, USA). Two independent observers evaluated lesion volume, maximum plaque burden, lesion CT number mean and standard deviation, and relative lesion composition. We found 10 noncalcified, 16 mixed, and 9 calcified lesions in our study cohort. Relative interobserver bias and variability for lesion volume were −37%, −13%, −49%, −44% and 28%, 16%, 37%, and 90% for all, noncalcified, mixed, and calcified lesions, respectively. Absolute interobserver bias and variability for relative lesion composition were 1.2%, 0.5%, 1.5%, 1.3% and 3.3%, 4.5%, 7.0%, and 4.4% for all, noncalcified, mixed, and calcified lesions, respectively. While mixed and calcified lesions demonstrated a high degree of lesion volume interobserver variability, noncalcified lesions had a lower degree of lesion volume interobserver variability. In addition, relative noncalcified lesion composition had a very low interobserver variability. Therefore, there may a role for MDCT in serial noncalcified plaque assessment with semiautomatic analysis software.

Three-dimensional thrombus segmentation in abdominal aortic aneurysms using graph search based on a triangular mesh

An abdominal aortic aneurysm (AAA) is the area of a localized widening of the abdominal aorta, with a frequent presence of thrombus. Segmentation and quantitative analysis of the thrombus in AAA are of paramount importance for diagnosis, risk assessment and determination of treatment options. The proposed thrombus segmentation method utilizes the power and flexibility of the 3-D graph search approach based on a triangular mesh. The method was tested in 9 3-D MDCT angiography data sets (9 patients with AAA, 1300 image slices), and the mean unsigned errors for the luminal and thrombotic surfaces were 0.99+/-0.18 mm and 1.90+/-0.72 mm. To achieve these results, 9.9+/-10.3 control points needed to be interactively entered on 2.1+/-2.2 image slices per 3-D CTA data set.

Simulation model for contrast agent dynamics in brain perfusion scans

Standardization efforts are currently under way to reduce the heterogeneity of quantitative brain perfusion methods. A brain perfusion simulation model is proposed to generate test data for an unbiased comparison of these methods. This model provides realistic simulated patient data and is independent of and different from any computational method. The flow of contrast agent solute and blood through cerebral vasculature with disease‐specific configurations is simulated. Blood and contrast agent dynamics are modeled as a combination of convection and diffusion in tubular networks. A combination of a cerebral arterial model and a microvascular model provides arterial‐input and time‐concentration curves for a wide range of flow and perfusion statuses. The model is configured to represent an embolic stroke in one middle cerebral artery territory and provides physiologically plausible vascular dispersion operators for major arteries and tissue contrast agent retention functions. These curves are fit to simpler template curves to allow the use of the simulation results in multiple validation studies. A γ‐variate function with fit parameters is proposed as the vascular dispersion operator, and a combination of a boxcar and exponential decay function is proposed as the retention function. Such physiologically plausible operators should be used to create test data that better assess the strengths and the weaknesses of various analysis methods. Magn Reson Med, 2010.

Quantification of aortic valve area at 256-slice computed tomography: Comparison with transesophageal echocardiography and cardiac catheterization in subjects with high-grade aortic valve stenosis prior to percutaneous valve replacement

PURPOSE The purpose of this study was to compare planimetric aortic valve area (AVA) measurements from 256-slice CT to those derived from transesophageal echocardiography (TEE) and cardiac catheterization in high-risk subjects with known high-grade calcified aortic stenosis. METHODS AND MATERIALS The study included 26 subjects (10 males, mean age: 79±6; range, 61-88 years). All subjects were clinically referred for aortic valve imaging prior to percutaneous aortic valve replacement from April 2008 to March 2009. Two radiologists, blinded to the results of TEE and cardiac catheterization, independently selected the systolic cardiac phase of maximum aortic valve area and independently performed manual CT AVA planimetry for all subjects. Repeated AVA measurements were made to establish CT intra- and interobserver repeatability. In addition, the image quality of the aortic valve was rated by both observers. Aortic valve calcification was also quantified. RESULTS All 26 subjects had a high-grade aortic valve stenosis (systolic opening area <1.0 cm(2)) via CT-based planimetry, with a mean AVA of 0.62±0.18. In four subjects, TEE planimetry was precluded due to severe aortic valve calcification, but CT-planimetry was successfully performed with a mean AVA of 0.46±0.23 cm(2). Mean aortic valve calcium mass score was 563.8±526.2 mg. Aortic valve area by CT was not correlated with aortic valve calcium mass score. A bias and limits of agreement among CT and TEE, CT and cardiac catheterization, and TEE and cardiac catheterization were -0.07 [-0.37 to 0.24], 0.03 [-0.49 to 0.55], 0.12 [-0.39 to 0.63]cm(2), respectively. Differences in AVA among CT and TEE or cardiac catheterization did not differ systematically over the range of measurements and were not correlated with aortic valve calcium mass score. CONCLUSION Planimetric aortic valve area measurements from 256-slice CT agree well with those derived from TEE and cardiac catheterization in high-risk subjects with known high-grade calcified aortic stenosis.

Fully automated segmentation of carotid and vertebral arteries from contrast enhanced CTA

We propose a method for segmenting and labeling the main head and neck vessels (common, internal, external carotid, vertebral) from a contrast enhanced computed tomography angiography (CTA) volume. First, an initial centerline of each vessel is extracted. Next, the vessels are segmented using 3D active objects initialized using the first step. Finally, the true centerline is identified by smoothly deforming it away from the segmented mask edges using a spline-snake. We focus particularly on the novel initial centerline extraction technique. It uses a locally adaptive front propagation algorithm that attempts to find the optimal path connecting the ends of the vessel, typically from the lowest image of the scan to the Circle of Willis in the brain. It uses a patient adapted anatomical model of the different vessels both to initialize and constrain this fast marching, thus eliminating the need for manual selection of seed points. The method is evaluated using data from multiple regions (USA, India, China, Israel) including a variety of scanners (10, 16, 40, 64-slice; Brilliance CT, Philips Healthcare, Cleveland, OH, USA), contrast agent dose, and image resolution. It is fully successful in over 90% of patients and only misses a single vessel in most remaining cases. We also demonstrate its robustness to metal and dental artifacts and anatomical variability. Total processing time is approximately two minutes with no user interaction, which dramatically improves the workflow over existing clinical software. It also reduces patient dose exposure by obviating the need to acquire an unenhanced scan for bone suppression as this can be done by applying the segmentation masks.

Automatic whole heart segmentation in CT images: method and validation

Deformable models have already been successfully applied to the semi-automatic segmentation of organs from medical images. We present an approach which enables the fully automatic segmentation of the heart from multi-slice computed tomography images. Compared to other approaches, we address the complete segmentation chain comprising both model initialization and adaptation. A multi-compartment mesh describing both atria, both ventricles, the myocardium around the left ventricle and the trunks of the great vessels is adapted to an image volume. The adaptation is performed in a coarse-to- fine manner by progressively relaxing constraints on the degrees of freedom of the allowed deformations. First, the mesh is translated to a rough estimate of the hearts center of mass. Then, the mesh is deformed under the action of image forces. We first constrain the space of deformations to parametric transformations, compensating for global misalignment of the model chambers. Finally, a deformable adaptation is performed to account for more local and subtle variations of the patients anatomy. The whole heart segmentation was quantitatively evaluated on 25 volume images and qualitatively validated on 42 clinical cases. Our approach was found to work fully automatically in 90% of cases with a mean surface- to-surface error clearly below 1.0 mm. Qualitatively, expert reviewers rated the overall segmentation quality as 4.2±0.7 on a 5-point scale.

3-D segmentation and quantitative analysis of inner and outer walls of thrombotic abdominal aortic aneurysms

An abdominal aortic aneurysm (AAA) is an area of a localized widening of the abdominal aorta, with a frequent presence of thrombus. A ruptured aneurysm can cause death due to severe internal bleeding. AAA thrombus segmentation and quantitative analysis are of paramount importance for diagnosis, risk assessment, and determination of treatment options. Until now, only a small number of methods for thrombus segmentation and analysis have been presented in the literature, either requiring substantial user interaction or exhibiting insufficient performance. We report a novel method offering minimal user interaction and high accuracy. Our thrombus segmentation method is composed of an initial automated luminal surface segmentation, followed by a cost function-based optimal segmentation of the inner and outer surfaces of the aortic wall. The approach utilizes the power and flexibility of the optimal triangle mesh-based 3-D graph search method, in which cost functions for thrombus inner and outer surfaces are based on gradient magnitudes. Sometimes local failures caused by image ambiguity occur, in which case several control points are used to guide the computer segmentation without the need to trace borders manually. Our method was tested in 9 MDCT image datasets (951 image slices). With the exception of a case in which the thrombus was highly eccentric, visually acceptable aortic lumen and thrombus segmentation results were achieved. No user interaction was used in 3 out of 8 datasets, and 7.80 ± 2.71 mouse clicks per case / 0.083 ± 0.035 mouse clicks per image slice were required in the remaining 5 datasets.

Three-dimensional and Four-dimensional Cardiopulmonary Image Analysis

Modern medical imaging equipment can provide data that describe the anatomy and function of structures in the body. Image segmentation techniques are needed to take this raw data and identify and delineate the relevant cardiovascular and pulmonary anatomy to put it into a form suitable for 3D and 4D modeling and simulation. These methods must be able to handle large multi-dimensional data sets, possibly limited in resolution, corrupted by noise and motion blur, and sometimes depicting unusual anatomy due to natural shape variation across the population or due to disease processes. This chapter describes modern techniques for robust, automatic image segmentation. Several applications in cardiovascular and pulmonary imaging are presented.